Parts-of-speech tagging is the process of labeling each word in a sentence. A tag mentions the word’s usage in the sentence. Usually, these tags indicate syntactic classification like noun or verb, and sometimes include additional information, with case markers (number, gender etc) and tense markers. A large number of current language processing systems use a parts-of-speech tagger for pre-processing. There are mainly two approaches usually followed in Parts of Speech Tagging. Those are Rule based Approach and Stochastic Approach. Rule based Approach use predefined handwritten rules. This is the oldest approach and it use lexicon or dictionary for reference. Stochastic Approach use probabilistic and statistical information to assign tag to words. It use large corpus, so that Time complexity and Space complexity is high whereas Rule base approach has less complexity for both Time and Space. Stochastic Approach is the widely used one nowadays because of its accuracy. Malayalam is a Dravidian family of languages, inflectional with suffixes with the root word forms. The currently used Algorithms are efficient Machine Learning Algorithms but these are not built for Malayalam. So it affects the accuracy of the result of Malayalam POS Tagging. My proposed Approach use Dictionary entries along with adjacent tag information. This algorithm use Multithreaded Technology. Here tagging done with the probability of the occurrence of the sentence structure along with the dictionary entry.

1. International Journal of Computer Science & Information Technology (IJCSIT) Vol 7, No 5, October 2015
DOI:10.5121/ijcsit.2015.7509 121
A NEW APPROACH TO PARTS OF SPEECH TAGGING
IN MALAYALAM
D. Muhammad Noorul Mubarak1
, Sareesh Madhu2
, S A Shanavas2
1
Department of Computer Science, University of Kerala, India
2
Department of Linguistics, University of Kerala, India
ABSTRACT
Parts-of-speech tagging is the process of labeling each word in a sentence. A tag mentions the word’s
usage in the sentence. Usually, these tags indicate syntactic classification like noun or verb, and sometimes
include additional information, with case markers (number, gender etc) and tense markers. A large number
of current language processing systems use a parts-of-speech tagger for pre-processing.
There are mainly two approaches usually followed in Parts of Speech Tagging. Those are Rule based
Approach and Stochastic Approach. Rule based Approach use predefined handwritten rules. This is the
oldest approach and it use lexicon or dictionary for reference. Stochastic Approach use probabilistic and
statistical information to assign tag to words. It use large corpus, so that Time complexity and Space
complexity is high whereas Rule base approach has less complexity for both Time and Space. Stochastic
Approach is the widely used one nowadays because of its accuracy.
Malayalam is a Dravidian family of languages, inflectional with suffixes with the root word forms. The
currently used Algorithms are efficient Machine Learning Algorithms but these are not built for
Malayalam. So it affects the accuracy of the result of Malayalam POS Tagging.
My proposed Approach use Dictionary entries along with adjacent tag information. This algorithm use
Multithreaded Technology. Here tagging done with the probability of the occurrence of the sentence
structure along with the dictionary entry.
KEYWORDS
NLP, POS tagger, Rule based approach, Stochastic approach, Multithreading, Dictionary entry,
Malayalam.
1. INTRODUCTION
Parts-of-speech tagging is simply a grammatical tagging. In natural language, there are a small
number parts of speech those are noun, adjective, adverb, verb, Conjunction, preposition etc.).
Words are grouped into different classes mentioned above. This type of grouping is called Parts
of Speech Tagging. Apart from the natural language, automated models have more numbers of
POS tags. Parts-of-speech tagging is the process of assigning a label to each and every word in a
corpus. There are two approaches used for automated Parts of Speech Tagging namely Rule based
and Stochastic. Rule based Approach follows predefined rule set whereas Stochastic Approach
follows Statistical measures. For this, Stochastic Approach use Machine Learning Algorithms like
Hidden Markov Model, Support Vector Machine etc.
Malayalam is a Dravidian family of languages, inflectional with suffixes with stem word. The
above mentioned Algorithms are efficient Machine Learning Algorithms but these are not built
for Malayalam. So it affects the accuracy of the result of Malayalam POS Tagging.

2. International Journal of Computer Science & Information Technology (IJCSIT) Vol 7, No 5, October 2015
122
This is an attempt to develop a new Algorithm that is specially designed for Malayalam POS
Tagging. This follows hybrid Approach, i.e. it use rule set of Malayalam with lexical dictionary
and Statistical measures based on the tagging structure.
2. TAGGING APPROACHES
There are two approaches in POS tagging: rule-based approach and stochastic approach. Rule-
based taggers use a large database of words with predefined disambiguation rule. For example,
that an ambiguous word is a noun or a verb. Stochastic taggers use training corpus along with
statistical algorithms to check for the ambiguity
RULE-BASED PARTS-OF-SPEECH TAGGING
Rule based approach is one of the oldest approach in tagging. It uses predefined rules. The
earliest algorithms for automatically assigning parts-of-speech were based on a two-stage
architecture. The first stage uses a dictionary to label word. The second stage use large lists of
predefined disambiguation rules to labelling word accurately .Rule based taggers use
morphological information along with predefined rule to assigns tags to unknown or ambiguous
words.
Advantages of Rule Based Taggers:-
a. use simple rules.
b. less storage.
Drawbacks of Rule Based Taggers:-
a. Generally less accurate as compared to stochastic taggers.
STOCHASTIC PARTS-OF-SPEECH TAGGING
Stochastic Approach use probability and statistical information for assigning tag to words. This
approach use statistical algorithms rather than grammar rule.
Advantages of Stochastic Parts of Speech Taggers:-
a. Generally more accurate as compared to rule based taggers.
Drawbacks of Stochastic Parts of Speech Taggers:-
a. Relatively complex.
b. Require vast amounts of stored information.
Stochastic taggers are popularly used as compared to rule based taggers because of their higher
degree of accuracy. On the other hand, this high amount of accuracy is achieved using some
relatively complex procedures and data structures.
3. OBJECTIVE
To develop a high accurate and low complexity (both time and space) Parts of Speech Tagger for
Malayalam. A Graphical User Interface is developed for entering input text in Malayalam and
also it shows tagged output in Malayalam. The input texts tokenize and analyzing
morphologically based on some rules and context, then tagged based on the lexical database.

3. International Journal of Computer Science & Information Technology (IJCSIT) Vol 7, No 5, October 2015
123
4. METHODOLOGIES
Parts of Speech Tagging developed here based on a new approach, we can termed it as hybrid
approach There are two types of POS tagging approaches-: Rule based and Stochastic based. First
one use predefined rule set and second one use probability measures. Compromising with both
approaches’ merits and demerits, we develop POS tagging algorithm that use lexicon dictionary
and structure of the sentence for evaluation or labeling the words.
The GUI designed using Netbeans(Java).
5. LITERATURE REVIEW
For language processing applications like Parser and Chunker, tagger with highest possible
accuracy is required. Usually Statistical approach gives better accuracy compared to Rule based
Approach. Parts-of-speech tagging is also a very practical application, with uses in many areas,
including machine translation, parsing, information retrieval and lexicography. Initially people
engineered rule for tagging, sometimes with the aid of a corpus. Later, with the aid of large
corpus, Markov-model based stochastic taggers that were trained automatically gives highly
accurate tagging result.
Recently, a number of approaches to POS tagging based on statistical and machine learning
techniques are applied, including among many others like Hidden Markov Models, and Support
Vector Machines. Many natural language tasks require the accurate assignment of Parts-Of-
Speech (POS) tags to previously unseen text. Due to the accessibility of huge corpus which has
been annotated manually with POS label, many taggers use annotated text to learn probability
rules and use them to tag without human intervention to unseen text.
6. OVER VIEW OF MALAYALAM LANGUAGE
Malayalam, the language categorized under the family of Dravidian languages spoken in Kerala.
This language uses around 37 million people. Liilaatilakam, is generally considered as the earliest
dissertation referring to grammatical structures of Malayalam
In the earlies of 19th centuary Malayalam did not have a proper grammar.
Malayalabhasaavyaakaranam published in 1851 by Hermann Gundert and the revised version
published in 1868 was the first proper grammatical dissertation of Malayalam.
Malayaalmayutevyaakaranam(1863) by Rev. George Mathen, Keeralabhaasaavyaakaranam by
PachuMootthatu, Keeralapaaniniyam A.R RajaraajaVarma and Vyaakaranamitram(1904) by M.
SeshagiriPrabhu followed. Grammatical literature from this point of time was fundamentally
paying attention on Keeralapaaniniiyam, which came to have almost the status of an authorized
grammar of Malayalam.
A regular grammatical custom illustration on a variety of grammars failed to develop and as a
result the framework of Keeralapaaniniiyam continued as the solitary grammatical model in
Malayalam. The grammars written in the post- Keeralapaaniniiyam period are basically
descriptive treatises on Keeralapaaniniiyam. While a few grammarians have suggested other
analyses in some areas, the grammars themselves truly follow the basic structure of
RajarajaVarma. For an era of more than 80 years from Keeralapaaniniiyam, no grammarian
attempted extend the Keeralapaaniniyam model to produce a more wide-ranging treatment of
Malayalam or to evaluate the grammatical structure of Malayalam using alternative models of
grammatical description. Keeralapaaniniyam and other traditional grammars have broadly

4. International Journal of Computer Science & Information Technology (IJCSIT) Vol 7, No 5, October 2015
124
covered the morphology of the language. However, there is little in them about syntax and
semantics. To deal with the formation of a modern language like Malayalam using a controlled
grammatical model has had severe repercussions in many fields.
Malayalam has 53 letters including 20 long and short vowels and others are termed as consonants.
Malayalam sentences
A sentence is a cluster of words that makes absolute logic. There are four types of sentences
available in Malayalam based on activities. Based on production they are of three types.
Sentence Classification Based on activities.
Malayalam sentences are basically of four types based on their activities.
• Assertive Sentence -This sentence makes a statement.
• Interrogative sentence -This ask a question.
• Imperative sentence -This show a command, appeal or a desire.
• Exclamatory Sentence -To express strong feeling, happiness, sorrow or wonder these kinds of
sentences are used.
Sentence Classification based on production
Sentences in Malayalam can be of three types based on the production
a. Simple
b. complex
c. Compound
• Simple sentences- This type contains only one main clause.
• Complex sentences-These sentences contain one primary clause and any number of secondary
clauses.
• Compound sentences -Compound sentences have any number of main clauses
7. TAGSET FOR MALAYALAM
This tag set has been developed based on the IIIT Hyderabad tag set for Indian languages.
It includes 28 tags.
Table 1.Tagset for Malayalam
Tag Name Type Tag Name Type
NN Noun RDP Reduplication
NST Noun denoting spatial and
temporal expressions.
CC Conjuncts( coordinatingand
subordinating)
NNP Proper Nouns UNK Unknown words
NNS Plural Nouns PRO Proverbs
PRP Pronoun IDM Idioms
DEM Demonstrative NNPC Compound Proper Nouns
VM Verb Main NLC Noun Locative

5. International Journal of Computer Science & Information Technology (IJCSIT) Vol 7, No 5, October 2015
125
VAUX Auxiliary Verb DOT
JJ Adjective QF Quantifiers
RB Adverb QC Cardinal
PSP Postposition QO Ordinal
ECH Echo words INTF Intensifier
WQ Question Words NEG Negation
SYM Special Symbol QM Question Mark
Here I am drop the Tag NPC(Noun Compound) from IIITH Tag set and Add a TAG as
NLC(Noun Locative). The tag NLC is necessary for most of the POS tagging Applications.
In my findings There is no need for tagging compound noun in Malayalam.
Consider the example:
In Malayalam this is a single word, there is no need to treat as compound noun.
Consider another example:
As the above example this is also a single word. Both these words can be dividing to two
morphemes but in Malayalam Dictionary, these two words are exists. As my Tagging follows
lexical dictionary, there is no need for NPC(Noun Compound) tag.
8. PROPOSED ALGORITHM FOR PARTS OF SPEECH TAGGING
Usually, for Parts of Speech Tagging, some statistical algorithms like Support Vector Machine,
Hidden Markov Model, etc. are used. These algorithms are powerful and efficient but it doesn’t
build for Parts of Speech Tagging in Malayalam Specifically. So that it affects the accuracy of
tagging process. In Dravidian languages, particularly for Malayalam language, inflected noun and
verb forms are common. Nouns may have inflected with plural marker and case marker. Verbs
might inflect with tense markers and are adjectivalized and adverbialized. So, many times we
need to depend on syntactic function or context to decide whether the word is a noun or adjective
or adverb or post position. This leads to the complexity in Malayalam POS tagging.
A noun may be categorized differently as common noun, proper noun or compound noun.
Likewise, verb may be finite, infinite or gerund. Other parts of speech were also divided into their
own subcategories.
For example, Malayalam word ‘paadilla’( ) in the following sentences gives different
parts of speech.
Avanpaadilla
Pukavalipaadilla
In the first sentence, the word ‘paadilla’ is a verb whereas in the second sentence it is a noun. This
is not rare in natural languages. A large amount of words are ambiguous. Also, the parts of speech
are many more POS tags rather than noun, verb and post position etc.

6. International Journal of Computer Science & Information Technology (IJCSIT) Vol 7, No 5, October 2015
126
So, here we develop a simple but powerful algorithm for Parts of Speech Tagging for Malayalam.
Here the Algorithm first performs tagging based on the lexical entries and then use statistical data
if necessary, i.e..suppose a word sometimes act as noun and some other time verb.
Proposed Algorithm for POS tagging
1 TAKE INPUT TEXT.
2 TOKENIZE THE INPUT TEXT (PRE-EDITING).
3 MANUAL TAGGING.
4 (First thread.) IF WORD EXIST IN DICTIONARY THEN
4.1 IF MULTIPLE TAG THEN
STORE IT IN A BUFFER
4.2 ELSE
PUT THE APPROPRIATE TAG FROM DICTIONARY.
REPEAT THESE STEPS UNTIL THE END.
5. (Second Thread) ELSE
1.1 DO STEMMING
1.1.1 IF CASE MARKERS EXIST
1.1.1.1 IF SUFFIX INCLUDES “KAL’ OR ‘MAAR’
TAG AS ‘NNS’(PLURAL NOUN)
1.1.1.2 ELSE IF SUFFIX INCLUDES ‘IL’ OR ’ILE’
TAG AS NLC (NOUN LOCATIVE)
5.1.1.3 ELSE
TAG AS ‘NN’ (NOUN)
1.1.2 IF TENSE MARKERS EXIST
TAG AS ‘VM’ (VERB MAIN)
6.(Third Thread) IF AN UNKNOWN WORD X
6.1 IF AN UNKNOWN WORD X IS PRECEDED BY A DETERMINER AND
FOLLOWED BY A NOUN,
TAG IT AS AN ADJECTIVE
6.2. IF (Prev_word is NOUN or PRONOUN) THEN
CURRENT WORD TAG AS ‘PSP’ (POSTPOSITION)
6.3. ELSE IF THE CURRENT_WORD’s SUFFIX INCLUDES ‘YUM’ OR ‘OO’
CURRENT_Word TAG AS ‘CC’ (CONJUNCTION)
6.4. ELSE IF (Prev_word is VERB) AND (Current_Word is VERB)
Current_Word TAG AS ‘VAUX (Auxiliary VERB)
7 IF THE NEXT WORD IS ADJECTIVE THEN
TAG CURRENT_WORD AS INTENSIFIER
8. (Fourth Thread)READ FROM BUFFER
IF (THE TAGGING STRUCTURES IN THE DATABASE (CORPUS) EQUALS
CURRENT WORD’S SENTENCE STRUCTURE ) THEN
PUT APPROPRIATE TAG
ELSE
ABANDON FOR MANUAL TAGGING
9. GET THE TAGGED OUTPUT TEXT.
10. INSERT THOSE NEW WORDS IN LEXICON.
The POS tags are identified by doing a lexicon lookup of the root word. This is especially useful
for morphologically rich Indian languages like Malayalam which inflect for gender, number, case
etc. The figure 1 below shows the various components of the POS tagger. An “ambiguity
resolver” is added to improve the accuracy of the POSTagger.

7. International Journal of Computer Science & Information Technology (IJCSIT) Vol 7, No 5, October 2015
127
Figure 1.Pictorial representation of the system
The architecture consist different modules based on their functionalities. The functionalities of
each of this module are explained briefly as follow.
Tokenize: Untagged sentences are downloaded from Malayalam newspapers and commercial
websites. We changed the input text into a column format suitable to the Algorithm. We used
blank space as the column separator. The corpus data was tokenized as the input data to the
algorithm must be in form of token.
Manual Tagging: The tokenizing module produces a corpus of untagged tokens. After which, the
corpus is tagged manually using proposed IIITH tag set. Initially around 20,000 words are tagged
manually.
SUFFIX SEPARATION RULES: The requirement of a pre-processing step in the training
phase is exclusively attributable to the unusual characteristics of Malayalam language. The
inflected word form in Malayalam can have multiple suffixes appended to its stem. This
characteristic of Malayalam language reduces the possibility of a word in the corpus to be present
in its stem. For example the word ‘ ‘ appears in the corpus in different forms, for
example, , etc..

8. International Journal of Computer Science & Information Technology (IJCSIT) Vol 7, No 5, October 2015
128
Fig 2: Suffix separation Malayalam
Noun Analysis
The general format of input and output of the morphological analyzer of Malayalam is as follows
Word -> stem + suffixes
Linguistic categorization of Nouns, which takes cases markers as Person, Number and Gender
information. The categorical information in noun is listed in Table 2 below.
Verb Analysis
Verb in natural language is a grammatical category, which includes tense along with it. Also
apprehensive information ie, tense, aspect and modularity (TAM) can be extracted from a verbal
form. Many markers are there in the TAM and it is listed in Table 3.
Apart from this two other categories that play an important role in the morphological analysis are
a. Negations
b. Linkmorph.
Malayalam has the following negative markers-: ”aatt”,”aat”,”NTa”

9. International Journal of Computer Science & Information Technology (IJCSIT) Vol 7, No 5, October 2015
129
Table 2: Noun Analysis
Table 3: Verb Analysis
Ambiguity Resolver
A word in the lexicon contains tags and tag information of the adjacent words. If the word
carrying single tag, then no complication, tag that word with the carrying tag. If the word carrying
multiple tags, i.e. an ambiguous word, the system looks for the current word’s adjacent tag
information and those information cross matches with the already stored tag information and tag
based on the tag structure. As the system follows multithreading technology, when it comes in the
fourth thread adjacent words should tagged. Following figure 4 shows the same graphically.
Table 4: Ambiguity Resolver
Word Tag Tag structure

10. International Journal of Computer Science & Information Technology (IJCSIT) Vol 7, No 5, October 2015
130
NN(Noun)
VM(Verb)
NN NLC VM
NN VM
9. CONCLUSION
Parts-of-Speech tagging, the assignment of Parts-of-Speech to the words in a given context of
use, is a basic technique in many systems that handle natural languages. Tags play an important
role in Natural language applications like text summarization, information retrieval and
information extraction etc. In order to alleviate problems for Malayalam language, we proposed a
new POS tagger approach. This paper describes a method for supervised training of a Parts-of-
Speech tagger using newly developed algorithm for Malayalam. We identified the ambiguities in
Malayalam lexical items, and developed a tag set appropriate for Malayalam. Finally, an efficient
and accurate POS Tagger model for Malayalam language is built. We hope this will be very
useful in natural language application like bilingual machine translation and in many areas.
REFERENCES
[1]. Dr. Ravi Sankar S Nair, A GRAMMAR OF MALAYALAM, Language in India (ISSN 1930-2940),
NOV 2012.
[2]. S. DeRose. “Grammatical category disambiguation by statistical optimization”.Computational
Linguistics, 1988, pp 14:31-39.
[3]. James Allen. Natural Language Understanding. Pearson Education, Second edition, 2002.
[4]. K. S. NarayanaPillai. AdhunikaMalayalaVyakaranam. The State Institute Of Languages,Kerala,
Thiruvananthapuram-3, Second edition,2003.
[5]. Vinod P M, Jayan V and Bhadran V K, Implementation of Malayalam Morphological Analyzer Based
on Hybrid Approach, in Proceedings of Twenty- Fourth Conference on Computational Linguistics
and Speech Processing,2012.
[6]. A.R.Raja Raja Varma.”Kerala Paaniniiyam”.ISDL, 1999.
[7]. [Brill 1992] E. Brill, A simple-rule based part-of-speech tagger, In Proceedings of the Third
Conference on Applied Natural Language Processing, Trento, Italy, 1992.